Mobile Tech Explained: Autonomous networks

This article is the first in a new series that aims to break down the complex systems and technologies behind the Rakuten Mobile network and Rakuten Communications Platform. In this edition, we speak with Dr. Petrit Nahi, Head of Data Science at Rakuten Mobile, Inc.

Ever since mobile networks were first deployed more than three decades ago, their infrastructure has required constant monitoring from hundreds or even thousands of engineers, each watching specific parameters and intervening as necessary. The downside: This kind of manual oversight is both expensive and error-prone — human beings make mistakes.

Now, with the rollout of 5G networks, things need to be done differently. With highly versatile and multi-faceted 5G networks serving an ever-expanding array of connected devices and machines, it is almost impossible for even a large team of engineers to absorb everything going on in any one moment — a commercial network generates billions of transactions or connections on a daily basis. Without a complete picture, there is a danger that changing one aspect of the network for the better will have an unintended consequence somewhere else.

5G networks are designed to be highly dynamic, supporting functionalities such as network slicing that will allow operators to deliver services customized to support particular customers and particular market segments. This means that traditional ways of managing networks are no longer feasible — automation and autonomy are a must.

A new age of automation

Fortunately, help is at hand. As computing becomes more readily available and easier to scale in cloud-native platforms, applications or systems can now digest the vast amounts of data being generated by and collected in a live mobile network.

The analytics on these data sets don’t just keep tabs on the performance of network components, they also provide clear insights into the customer experience — how fast and responsive the connectivity appears to be to the end-user. By combining data from many different sources and applying artificial intelligence techniques, these newly-developed applications or systems generate a comprehensive picture of what is going on in the network (as perceived by the end user) and yield new insights that would have been hidden from engineers in the past.

5G networks are designed to be highly dynamic. This means that traditional ways of managing networks are no longer feasible — automation and autonomy are a must.

These kinds of systems are being deployed across the economy: Factories and even whole supply chains are using real-time data to become increasingly automated and efficient. Advances in artificial intelligence are helping to make these automated systems even smarter. Machine learning, in which software learns to recognize patterns in real-world data, is being applied successfully in many different fields – from transcribing speech accurately and identifying the subject of a photograph to enabling vehicles to drive themselves.

Rakuten Mobile’s smart network

Rakuten Mobile uses “self-organizing network” solutions to help optimize the end-to-end performance of its network in Japan. These systems make extensive use of advanced knowledge and data-driven algorithms to detect anomalies from the network performance point of view.

Rakuten Mobile goes way beyond the traditional concept of self-organizing networks by combining multiple data sources, advanced artificial intelligence techniques and the knowledge of its engineers to not only detect but also identify the root causes of any anomalies and recommend steps for their resolution.

Dr. Petrit Nahi, Head of Data Science at Rakuten Mobile
Dr. Petrit Nahi, Head of Data Science at Rakuten Mobile, Inc.

For example, a near real-time “inter-cell interference cancellation function” provided by Airhop Communications is used for something called “interference mitigation.” This allows Rakuten Mobile to optimize its network and significantly improve the user experience. The data generated is then combined with other network data to optimize the network. Software components deployed at each base station measure the level of interference, which can effect the quality of calls and other activity, with other base stations. Then, if necessary, the radio frequency transmitting the signal can be changed, just as a driver changes lanes on a highway to avoid other vehicles.

The result: An automated antenna tilt optimization algorithm powered by machine learning, which minimizes interference and improves network coverage while at the same time making the network highly available to the end user. In a somewhat similar way that solar panels can be optimized to follow the sun, Rakuten Mobile’s antennas leverage data and machine learning to automatically tilt and optimize the network.

Rakuten Mobile’s open RAN architecture and cloud-native deployment with readily available network and other data enables the introduction of intelligence like this into the network in a seamless fashion.

Predicting what will happen next

As Rakuten Mobile’s new network is evolving very quickly, the algorithms used in these systems also need to evolve — statistical and machine learning systems process actual data from the network and refine the algorithms over time.  As they detect more and more patterns in the data, these systems can go beyond detecting and help to predict what will happen next.

“We think we are heading towards a level 4 autonomous network … a network that is self-organized, self-optimized to address real-time issues that might happen in the underlying infrastructure.”

Tareq Amin, CTO of Rakuten Mobile, Inc.

In the event of network congestion or degradation (decreasing connectivity or response speed), “We can monitor who’s connected and who’s impacted. You can also look at the traffic patterns, or subscriber behavior and associated mobility,  and determine if the outage lasts for different amounts of time and how many others will be impacted by it,” explains Dr Petrit Nahi, Head of Data Science at Rakuten Mobile. “The network learns to adapt and compensate for any network degradation by adjusting how available resources are used.”

The crucial role of the cloud

All of this is possible because Rakuten Mobile’s network architecture is much more flexible, malleable and open than that of a traditional network. Crucially, all of its software runs on the cloud, where it can be quickly reconfigured and isn’t dependent on specialist hardware. This cloud-native and virtualized architecture makes it easier for Rakuten Mobile’s intelligent systems to access the necessary data to see what is going on in different elements of the network and adjust parameters where necessary. This architecture “tremendously simplifies the whole [network] and makes the approach possible,” notes Nahi. 

All the data in Rakuten Mobile’s network is aggregated in major data centers, where it can be quickly analyzed. As a result, the dashboards in Rakuten Mobile’s network operations center show both the current state of the network and how that is likely to change moving forward.

Thanks to this streamlined and flexible architecture, Rakuten Mobile’s network, which currently includes more than 9,000 base stations, is managed by a team of just 130 engineers — far less than traditional telcos. And while Rakuten Mobile is scaling its network and plans to cover more than 96% of the Japan population by summer 2021, the engineering team will remain relatively small.

Self-learning and adaptation: Toward full autonomy

Today, Rakuten Mobile largely relies on a hybrid model — a mix of human and automated oversight. But it is now looking to deploy fully autonomous systems in parts of the network where actions are relatively easy to define.  A major step on the path to autonomy is the introduction of “self-learning” in the network and, with it, adaptation. As the technology advances, Nahi expects these systems to take on greater responsibility, thus providing more reliability and quality to the network’s users.

Indeed, Rakuten Mobile CTO Tareq Amin has said his company’s mobile network architecture could achieve “level 4” automation — which means the network could operate without manual intervention, if need be — as early as 2022.

“We think automation is the underpinning of everything we are going to do in the network,” Amin told the TM Forum’s Digital Transformation World event in October of this year. “Key for anything in 5G, and even 4G, is what you do in automation …We think we are heading towards a level 4 autonomous network. In two years from today, I really believe this will no longer be a white paper… a network that is self-organized, self-optimized to address real-time issues that might happen in the underlying infrastructure.”

As it leverages the power of autonomous networks, Rakuten Mobile’s systems will become more and more efficient and effective. The end result: more cost-effective connectivity and a better end-user experience.

Show More
Back to top button